how can i custom more parameter of DeepSeek-R1model provided with DeepSeek,Siliconflow etc.?

[BeWater799]

What specific problem does this solve?

there are more useful params that can actually influence the efficience of agents, like "thingking-budget" make model output its thinking content in a limited length, which make this smart-and-economical model work response faster.

How should this be solved?

provide a window edits them when custom "model provider"

How will we know it works? (Acceptance Criteria)

its support by deepseek official api and other popular api provider like SiliconFlow and Volcengine-ark, which deeply colabrate with Deepseek and have huge groups of computing power, supporting OpenAI compitable API

Estimated effort and complexity

easy, i think.

Technical considerations (optional but helpful)

No response

Trade-offs and risks (optional)

No response

Additional context (optional)

No response

Proposal checklist

• I've searched existing Issues and Discussions for duplicates
• This is a specific, actionable proposal with clear problem and solution
• I've included concrete acceptance criteria
• I understand this needs approval before implementation begins

Interested in implementing this?

• Yes, I'd like to help implement this feature

[daniel-lxs]

Hey @BeWater799, which parameters are you referring to exactly?

[BeWater799]

sorry, ”thinking_budget”, actually.

[daniel-lxs]

We currently do support changing the thinking budget for models that support it, I wasn't able to find any information about Deepseek R1 supporting this parameter, do you have a source for it?

[BeWater799]

We currently do support changing the thinking budget for models that support it, I wasn't able to find any information about Deepseek R1 supporting this parameter, do you have a source for it?

just search on Siliconflow‘s official API doc for it,which is a big-popular provider for Qwen and DeepSeek Series' models, here's the doc's url of their site operating in China:

https://docs.siliconflow.cn/cn/api-reference/chat-completions/chat-completions
I thought there must be a international site, i couldnt look for it till my VPN traffic usage package is refreshed 3 dayz later, hope you find info useful before that, message me here if you need anything more!

[olweraltuve]

Hi folks — quick data point on max_tokens for DeepSeek models (DeepSeek provider).

According to the latest DeepSeek docs:

• deepseek-reasoner (R1-0528): max_tokens default = 32 K, maximum = 64 K – see the Reasoning-Model guide.
  https://api-docs.deepseek.com/guides/reasoning_model
• Pricing page also lists 64 K context length and 64 K max output for this model.
  https://api-docs.deepseek.com/quick_start/pricing

RooCode currently caps the setting at 8 K, so we can’t reach the model’s full capacity.

Proposal (happy to open a PR)

1. Add a range slider and an optional numeric input that let users choose any max_tokens value up to the provider’s true limit (64 000 for DeepSeek).
2. Pre-populate the field with the provider’s documented default (32 000 for deepseek-reasoner).

I’m available to implement and test this change — just let me know if this approach works for you. 👍

[olweraltuve]

@daniel-lxs